xlm-roberta-email-classifier

Fine-tuned version of xlm-roberta-base for multi-class classification of English-language emails.
This model is designed to automatically route or tag incoming messages based on their content.

Model Overview

• Base Model: xlm-roberta-base
• Task: Email classification (10 categories)
• Language: English
• Frameworks: Hugging Face Transformers, PyTorch Lightning
• Training Tracker: Weights & Biases

Performance

• Accuracy: 0.42
• F1 Score: 0.436
• Precision: 0.527
• Recall: 0.42

Class Labels

The model predicts one of the following categories:

Label ID	Category
0	Billing and Payments
1	Customer Service
2	General Inquiry
3	Human Resources
4	IT Support
5	Product Support
6	Returns and Exchanges
7	Sales and Pre-Sales
8	Service Outages and Maintenance
9	Technical Support

Usage

from transformers import AutoTokenizer, AutoModelForSequenceClassification

tokenizer = AutoTokenizer.from_pretrained("ale-dp/xlm-roberta-email-classifier")
model = AutoModelForSequenceClassification.from_pretrained("ale-dp/xlm-roberta-email-classifier")

email_text = "I'd like to return the item I purchased last week."
inputs = tokenizer(email_text, return_tensors="pt")
outputs = model(**inputs)

predicted_class_id = outputs.logits.argmax().item()
label_map = {
    'Billing and Payments': 0,
    'Customer Service': 1,
    'General Inquiry': 2,
    'Human Resources': 3,
    'IT Support': 4,
    'Product Support': 5,
    'Returns and Exchanges': 6,
    'Sales and Pre-Sales': 7,
    'Service Outages and Maintenance': 8,
    'Technical Support': 9
}
predicted_label = list(label_map.keys())[list(label_map.values()).index(predicted_class_id)]
print(predicted_label)